Jump to content


Recommended Posts

How Hurricane Harvey Became So Destructive




Scientists say the effects of Hurricane Harvey, which has been stalled over the Texas Gulf Coast since Friday and dumped more than 20 inches of rain in some areas, were worsened by a lethal confluence of meteorological events: warm water in the Gulf of Mexico that intensified the rainfall, and a lack of winds in the upper atmosphere that could have steered Harvey away from land.

Exacerbating the situation, said Hal Needham, a storm surge expert and founder of the private firm Marine Weather & Climate in Galveston, Tex., was that the storm surge elevated Galveston Bay, blocking drainage of the rain that pummeled coastal and inland areas.

“A two- or three-foot storm surge alone would not have been catastrophic,” Mr. Needham said. “It was all these ingredients coming together.”

And it’s not over.

Dennis Feltgen, a spokesman for the National Oceanic and Atmospheric Administration’s National Hurricane Center in Miami, said the driving rains would continue for another two or three days, pouring an additional 15 to 25 inches over parts of Southeast Texas. Some areas, he said, could see as much as 50 inches of rain.

“This is unprecedented,” he said.

Yet it does have some parallels. J. Marshall Shepherd, director of the atmospheric sciences program at the University of Georgia, said Harvey is very much like Allison, a tropical storm that flooded Houston badly in 2001 because it lingered over the city and dumped prodigious amounts of rain.

“In some ways, I think this event is going to far surpass what we saw in Tropical Storm Allison,” Dr. Shepherd said. The atmosphere is not helping to push it anywhere else. “The steering currents that would normally lift it out of that region aren’t there,” he said.

Hurricanes are essentially large weather engines fueled by the warm waters of the ocean below.

The mind-boggling amount of rainfall during Harvey is a function of the storm sitting by the Gulf of Mexico and continuing to draw moisture directly from it. Because of the orientation of the storm, Dr. Shepherd said, “you’ve just got this stream of moisture firehosing into the Houston region,” as the moisture is constantly replenished by the gulf. “This could go down as the worst flood disaster in U.S. history.”

Scientists are increasingly able to link some extreme weather events to climate change, but when it comes to hurricanes, many say there remain a number of unknowns. What is clear, though, is that rising global temperatures warm the oceans, which causes more water to evaporate into the atmosphere.

The buildup of moisture in turn contributes to the global increase in extreme rainfall, Kenneth Kunkel, a researcher with the North Carolina Institute for Climate Studies, said.

Even without climate change as a factor, Dr. Kunkel said, oceans are normally warm this time of year. But, he pointed out, the Gulf of Mexico has been warmer than average lately, most likely feeding into the deluge.

Several scientists stressed that while the damage of Hurricane Harvey was unrelenting, it was not unexpected.

Forecasters were anticipating a very wet storm to park over Houston for an extended period. “This, honestly, is playing out, unfortunately, exactly as we thought it would several days ago,” Dr. Shepherd said.

He said he grew worried when he saw public statements of relief that the storm had been downgraded from Category 4 strength to a tropical storm. He said the change in technical terminology may have confused the public and led local officials to lose focus on the greater threat, a multiday rain event.

“There was always a one-two punch with this particular storm, but we were always more concerned about the ‘two,’ the rainfall,” Dr. Shepherd said. “Once that subsided, people like me said, ‘O.K., we’re just getting started.’”

Different models predict different aspects of storms. Rick Luettich is director of the University of North Carolina’s Institute of Marine Sciences and an internationally recognized expert on storm surge, and a principal developer of the ADCIRC computer programs, which can be used to predict storm surge. For Harvey, Friday’s calculations suggested a surge of eight to 10 feet in the coastal areas, with as much as 12 feet in some of the shallow coastal estuaries.

The water threat that is getting the most attention with Harvey, however, is the intense rainfall predicted as the storm lingers over the Houston area. With storms growing wetter thanks to climate change, Mr. Luettich and his collaborators are trying to add rainfall calculations to the coastal surge forecasting model.

“People don’t care, if they got wet or got drowned, whether their water was salty or fresh,” he said.



  • Like 1
Link to comment

Hurricane Harvey provides lab for U.S. forecast experiments

By Paul Voosen Aug. 28, 2017 , 4:24 PM

Science Mag


For years, U.S. forecasters have envied their colleagues at the European Centre for Medium-Range Weather Forecasts (ECMWF) in Reading, U.K., whose hurricane prediction models remain the gold standard. Infamously, the National Weather Service (NWS) in 2012 failed to predict Hurricane Sandy’s turn into New Jersey, whereas ECMWF was spot on. But two innovations tested during Hurricane Harvey, one from NASA and another from the National Oceanic and Atmospheric Administration (NOAA), could help level the playing field.

NOAA’s offering is a brand-new forecasting model. Two years ago, NOAA’s Geophysical Fluid Dynamics Laboratory (GFDL) in Princeton, New Jersey, won a competition to provide the computer code for the next-generation weather model of NWS. Current NWS models must wait for results from a time-consuming global simulation before they can zoom in on a smaller area and run a high-resolution model for hurricanes. With GFDL’s new code, the next-generation model will be able to simulate storms at the same time as it runs globally, in theory, improving forecasts for hurricane paths because its fine-scaled predictions feed immediately into the model’s next run, rather than lagging behind.

Last week, GFDL anxiously watched the developing storm to see how it compared with a test run of the next-generation model. On Thursday, a day prior to landfall, the experiment agreed with the European model that Harvey would plow inland, stall, then head back out over the Gulf of Mexico before making a second landfall near Houston, Texas. That progression, close to what’s happening, helps explain the sustained, catastrophic rainfall that has battered the Texas coast.

The GFDL model, called FV3, also correctly forecasted that Harvey would develop a double eyewall—a second circular band of storms around the band enclosing the eye. The model’s zoomed-in view also predicted the extreme rainfall totals seen by Houston some 5 days in advance, says Shian-Jiann Lin, the GFDL scientist who led the development of the code powering FV3.

Intensity can be even harder to predict than storm paths, and here NASA may be able to help. Many models missed that Harvey would grow to a category-4 storm just prior to landfall, in part because data on wind speeds are spotty and difficult to collect. Last December, NASA launched a constellation of eight identical microsatellites, called the Cyclone Global Navigation Satellite System (CYGNSS), to fill the gap. CYGNSS works by detecting the surface roughness of the ocean—a proxy for wind speeds—from the reflected radio signals of GPS satellites. These long-wavelength signals can pass through the veil of rain that cloaks hurricanes and blocks the microwaves that traditional weather satellites detect.

Harvey was the first test for CYGNSS in severe winds. On 25 August, before Harvey made landfall, Christopher Ruf, an atmospheric scientist and engineer at the University of Michigan in Ann Arbor, strapped himself into a P-3 turboprop, an NOAA hurricane hunter, bound for the storm’s eye. His seat fell out beneath him again and again as the aircraft repeatedly plunged into the eyewall. Each time the wind grew more severe: The storm was rapidly intensifying.

It will take weeks to know whether CYGNSS captured this sharp intensification, Ruf says. The weather service will be following his results closely. The constellation is technically only a 2-year experiment, but it’s possible the satellites could be pressed into operational service for NOAA, Ruf says. “Our simulations have shown that the forecast skill is improved. Now we need to demonstrate it for real.”

  • Like 1
Link to comment

Sunspot-hurricane link proposed

Controversial research hints that solar cycle affects cyclone intensity.

Jeff Tollefson 2008 Sep 28




A new study suggests that more sunspots mean less intense hurricanes on Earth. But many hurricane experts are cool on the idea.


James Elsner, a climatologist at Florida State University in Tallahassee, has analyzed hurricane data going back more than a century. He says he has identified a 10- to 12-year cycle in hurricane records that corresponds to the solar cycle, in which the Sun's magnetic activity rises and falls.

news.2008.1136.jpgSolar activity varies on a roughly 11-year cycle, in which its magnetic activity waxes and wanes.NASA/TRACE

The idea is that increased solar activity - associated with sunspots - means more ultraviolet radiation reaching the Earth's upper atmosphere. That warms the airs aloft and decreases the temperature differential between high and low elevations that otherwise would fuel hurricanes.

"Our results indicate that there is an effect in the intensity of storms due to the higher temperatures aloft," says Elsner, who published the results on 19 September in Geophysical Research Letters1.

He says the statistical analysis suggests a 10% decrease in hurricane intensity for every 100 sunspots. At the peak of its cycle, the Sun might exhibit around 250 sunspots.

Hot debate

Establishing such a relationship would be enormously valuable, providing researchers, meteorologists and insurance companies with another tool for predicting storms and assessing financial liabilities. Several hurricane experts called the study provocative but raised questions about statistical analysis, as well as the physical processes at play.

Ka-Kit Tung, a climate researcher at the University of Washington in Seattle who has studied the solar cycle, says ultraviolet rays do most of their heating when they are absorbed by ozone higher in the stratosphere. But hurricanes don't make it past the tropopause, the boundary layer between the troposphere and the stratosphere that is about 16 kilometres up at storm latitudes. "We have not established that there is heating at the tropopause due to the solar cycle," Tung says.

For the new study, Elsner and a post-doctoral student, Thomas Jagger, used more than a century of records on hurricanes that reached land in the United States as a proxy for records on hurricane intensity. The assumption is that the effect on overall hurricane intensity would make itself evident as more or fewer hurricane-strength storms made land. But not everybody buys the idea that frequency can be reliably translated into intensity.

"This is something worth investigating, but they made too many assumptions for me to just accept their conclusion at this point," says Judy Curry, a hurricane researcher at the Georgia Institute of Technology in Atlanta. She and others questioned whether the influence of solar cycles would be strong enough to appear in such data sets.

In search of a mechanism

Earlier this month Elsner published a paper in Nature2 providing more evidence that rising ocean temperatures will increase the intensity of major storms around the globe. He says the solar-cycle pattern emerged after his team had already taken into account major factors such as sea-surface temperatures.

The team also found a correlation when comparing sunspot records with daily intensity values for storms dating back to 1944, using "best-track" records from the National Hurricane Center. "It gives us some faith that we are probably on to something here," Elsner says.

The effect would be felt in areas where sea-surface temperatures are high enough to produce big storms, he says. In marginal areas, where waters are cooler, solar activity could promote storms by providing the initial heat to seed a hurricane.

"It's an interesting statistical finding, but I'm unclear on the mechanisms here," says Thomas Knutson, a hurricane researcher with the US National Oceanic and Atmospheric Administration in Princeton, New Jersey. "There's more work needed to see if there really is a physical effect and to flesh out the mechanism." 

  • References

    1. Elsner, J. B. & Jagger, T. J. Geophys. Res. Lett. 35, L18705, doi:10.1029/2008GL034431 (2008).
    2. Elsner, J., Kossin, J. P. & Jagger, T. H. Nature 445, 92–95 (2008).
Link to comment

Evidence linking solar variability with US hurricanes


Robert E Hodges & James B. Elsner

First published: 14 July 2010

International Journal of Climatology



The relationship between US hurricanes and solar activity is investigated empirically. First, a relationship between the probability of a US hurricane and the solar cycle is shown conditional on sea surface temperatures (SST). For years of above normal SST, the probability of three or more US hurricanes decreases from 40 to 20% as sunspot numbers (SSN) increase from lower to upper quartile amounts. Second, since SST is in phase with the 11-year total solar irradiance cycle but upper-air temperature is in phase with ultraviolet radiation changes on the monthly time scale, an anomaly index of SSN is constructed. The index is significantly correlated with US hurricanes and major US hurricanes over the period 1866-2008. The chances of at least one hurricane affecting the United States in the lowest and highest SSN anomaly seasons are 68 and 91%, respectively. A similar relationship is noted using hurricane records spanning the period 1749-1850, providing independent corroborating evidence linking solar variability to the probability of a US hurricane. Copyright © 2010 Royal Meteorological Society

1. Introduction

Hurricanes, capable of filling the entire vertical extent of the troposphere, transfer large amounts of energy from the ocean to the atmosphere via latent heat released from the condensation of rising moist air. But their development and intensification are highly dependent on ambient influences. Changes in atmospheric water vapour composition (Emanuel, 1991), steering currents (Dong and Neumann, 1986), and sea surface temperatures (SST) (Holland, 1997) affect tropical cyclone intensity (for an overview, see Wang and Wu, 2004). Recently, it has been found that variations in upper-air temperature resulting from fluctuations in ultraviolet (UV) radiation from the sun can also affect tropical cyclone intensity (Elsner and Jagger, 2008 (hereafter, EJ08)).

Actually, the idea of a solar connection to tropical cyclones is almost a century and a half old (Meldrum, 1872; Poey, 1873), but the relationship has recently been re-examined. EJ08 show that the annual US hurricane counts are significantly related to solar activity. The relationship results from fewer intense tropical cyclones over the Caribbean and Gulf of Mexico when there are many sunspots. The finding is in accordance with the heat-engine theory of hurricanes, which predicts a reduction in the maximum potential intensity in response to warming in the atmospheric layer near the top of the hurricane. An active sun warms the lower stratosphere and upper troposphere through ozone absorption of additional UV radiation. In short, increased solar activity—associated with more sunspots—means more UV radiation reaching the Earth's upper atmosphere. The extra radiation warms the air aloft and decreases the temperature differential between high and low elevations resulting in less available potential energy for hurricane intensification (Elsner et al., 2010). But increased solar activity also contributes to warming of the ocean and altering of the atmospheric circulations, thus complicating the role solar variability plays in modulating hurricane activity.

The purpose of this paper is to (1) further examine the evidence for a sun–US hurricane relationship on the interannual time scale, and (2) show that the relationship is detectable in data that pre-dates the modern record. In doing so, the study provides additional evidence that the relationship is likely the result of changes in upper-level temperature caused by variation in UV radiation. This treatment is statistical, and no attempt is made to directly address the hypothesized causal mechanisms, as is done in Elsner et al. (2010).

The paper begins in Section 2 with background information about solar variability and its possible relationship with hurricane activity. In Section 3, the data used in the study are described. In Section 4, we examine the sun–hurricane relationship using data since 1851. The intra-seasonal variability in solar activity is proposed as a better way to describe the sun–hurricane relationship. This is shown using a table of ranked years and, more comprehensively, with a Poisson regression model. In Section 5, the technique of ranking years according to the magnitude of the change in intra-seasonal solar activity is used on a set of US hurricane counts dating back to 1749. The results are consistent in showing an increase in US hurricane probability for years with large sunspot number (SSN) anomalies, featuring relatively low numbers of sunspots during the hurricane season and high numbers immediately before and after the season.

2. Background

Sunspots are visible disturbances on the surface of the sun resulting from intense magnetic fields associated with the geographic switching of magnetic poles which occurs, on average, every 11 years (Schwabe cycle) as part of the solar magnetic cycle. Higher outgoing total solar irradiance (TSI) occurs during periods of increased SSN. While TSI in a given cycle varies by only 0.1% in magnitude for all radiation wavelengths, UV and extreme UV (EUV) have been shown to vary by more than 10% (Hoyt and Schatten, 1997). Increases in upper-tropospheric (above 200 hPa) temperature have been shown to occur during solar activity maxima (Labitzke, 2002; Pap and Fox, 2004), with tropical tropopause temperatures also increasing with increased UV radiation (Hood, 2003). Higher solar activity has also been linked with increased global SST. Upper ocean temperatures were shown to increase during periods of increased solar activity over most of the tropical oceans (White et al., 1997). In fact, Elsner et al. (2008) develop a model for North Atlantic SST that includes a term for the 11-year cycle.

The maximum potential intensity (MPI) theory postulates that, in an idealized thermodynamic environment, a hurricane's intensity is proportional to the thermodynamic efficiency between input energy (entropy gain from ocean/atmosphere interaction) and outflow venting (mechanical dissipation) (Miller, 1958; Emanuel, 1991; Holland, 1997; Bister and Emanuel, 1998). An increase (decrease) in surface (upper-tropospheric) temperature would increase the thermodynamic potential energy available for convection, all else being equal. The effect of upper-tropospheric and oceanic warming from increased UV exposure presents a bicameral climate effect upon hurricane intensity: warming at the surface increases a hurricane's potential intensity, whereas warming aloft decreases the potential mechanical dissipation and therefore hurricane intensity.

An active sun (many sunspots) might lead to warmer SST over regions where hurricanes form (due to additional shortwave energy absorbed by the surface). But it will more immediately lead to a warmer upper troposphere over these same regions (due to additional UV radiation absorbed by ozone). Since, according to the heat-engine theory, warming at the surface acts to enhance cyclone intensification while warming aloft acts to inhibit intensification, the influence of the sun on hurricanes is masked when using simple analyses. A way around this problem is to control for one or the other of the variables.

EJ08 do this geographically by examining the sun–hurricane relationship only for hurricanes over the western part of the tropical North Atlantic basin. Over this region, ocean heat content is nearly always high enough for tropical cyclogenesis, so the limiting thermodynamic factor is upper tropospheric temperatures (UTT). Indeed, they find a significant negative relationship between daily SSN and hurricane intensity over this part of the North Atlantic. Because of the region's already sufficiently warm ocean, the extra energy from an active sun limits intensification via a warmer upper troposphere. Thus EJ08 are able to detect a solar signal on the hurricane occurrence by geographically controlling for SST.

Over the western Atlantic including the Caribbean Sea and Gulf of Mexico where optimal oceanic heat content for cyclogenesis is spatially and intra-seasonally expansive, the limiting thermodynamic factor for a tropical cyclone to reach maximum intensity is the temperature of the atmosphere near the tropopause. Since this covariate is inversely related to MPI, we note that an active sun warms the lower stratosphere, thereby decreasing potential intensity. Indeed, the correlation between seasonally averaged SSN and temperatures near the tropopause over the domain is consistently positive based on atmospheric data over the period 1948–2006 (EJ08). In contrast, the daily SSN is positively correlated with daily averaged tropical cyclone intensity for cyclones over the eastern Atlantic.

The present work examines the sun–hurricane relationship by constructing an index that accounts for the low- and high-frequency variability inherent in the solar cycle. Though findings here corroborate those of EJ08, a potential bicameral role of the sun is examined through the construction of an SSN anomaly index of in-season and out-of-season values. The inclusion of almost 100 years of additional North Atlantic hurricane records back to 1749 (Chenoweth, 2006) confirms the robustness of the interannual sun–hurricane relationship.

3. Data

The National Hurricane Center (NHC) houses the North Atlantic-basin hurricane database (HURDAT, or Best Track), containing dates, tracks, wind speeds, and minimum central pressure values as available, providing the best available modern hurricane information dating back to 1851 (Landsea et al., 2004). With refining work begun by Jose Fernandez-Partagás (Partagás and Diaz, 1996), over 5000 additions and alterations have been approved by the NHC Best Track Change Committee in an effort to improve the accuracy of storm data between 1851 and 1910 (AOML-NOAA, 2008). Through the 2008 hurricane season, HURDAT identifies 1362 unique North Atlantic storms (>18 m s−1) and 819 individual hurricanes (>33 m s−1), 283 of which struck the US mainland.

Monthly mean SSN for this study are the International Sunspot Number as made available by the National Geographic Data Center, which was originally constructed by Solar Influences Data Analysis Center, World Data Center, at the Royal Observatory of Belgium (Van der Linden, 2009). Reliable monthly observations extend back to 1749. The Swiss astronomer Johann Rudolph Wolf introduced a daily measurement technique that observes both total spots and the quantity of their clusterings. The dataset addresses the observed error by incorporating a weighted average of cooperating observations.

Other covariate data include the North Atlantic oscillation (NAO), North Atlantic SST, and the Southern Oscillation index (SOI). The NAO May to June monthly index values (mean of − 0.35 and standard deviation of 1.02) are courtesy of the Climatic Research Unit, and are calculated from sea level pressures at Gibraltar and at a station over southwest Iceland (Jones et al., 1997). More information about the NAO as a covariate for US hurricane activity is provided in Elsner and Jagger (2006). Monthly SST anomalies averaged from June to November (mean of 0.01 °C and standard deviation of 0.20 °C) are departures from the long-term average for 0° to 70°N, and are linearly detrended. The SST data are provided and maintained by the Physical Sciences Division of the Earth System Research Laboratory, NOAA, and are derived from the Kaplan SST dataset (Kaplan et al., 1998). The SOI averaged from June to November (mean of − 0.09 and standard deviation of 0.92) is defined as the normalized sea-level pressure difference between Tahiti and Darwin, and is obtained also from the Climatic Research Unit.

A record of US hurricanes pre-dating those listed in HURDAT are available in Chenoweth (2006). This cross-referenced compendium of North Atlantic hurricanes improves upon previous compilations (Tannehill, 1938; Poey, 1855; Ludlum, 1963; Redfield, 1863), providing individual storm positions and dates through cross-referencing and validating individual entries using newspaper accounts, weather diaries, and ships logbooks for the period of 1700–1850. Chenoweth (2006) identifies 383 unique storms and 289 hurricanes, 127 of which affected the US mainland. The archive was digitized for use as a database of individual storm locations in Scheitlin et al. (2010) and we use this digitization here.

In summary, the present study considers the evidence for a sun–hurricane relationship by dividing the data into pre- and post-1851. The post-1851 data use the HURDAT records for US hurricane counts and the covariate information as described above. However, the monthly SST data begin in 1856 and the monthly SOI data begin in 1866, so adjustments are made accordingly. The pre-1851 analysis uses the Chenoweth (2006) archive for US hurricane counts and the covariate information is limited to monthly SSN.

4. The sun–hurricane relationship since 1851

4.1. Seasonal variability

As previously mentioned, one of the purposes of the present work is to re-examine the evidence shown in EJ08 for a sun–hurricane relationship. EJ08 assume a Poisson model for US hurricane counts and the model includes SST and SSN as covariates. Here, we make the same Poisson assumption and look at the bivariate relationship between US hurricanes and the SSN conditional on above- and below-normal SST.

Figure 1 shows the probability distribution of US hurricanes conditioned on values of August to October SSN for seasons of above-normal and below-normal SST, based on upper and lower tercile SST values averaged from June to November, respectively. For each hurricane count, the percent bar represents the probability of that many US hurricanes given the data and the Poisson distribution. For above-normal SST values among lowest terciled August to October SSN, the probability of no US hurricanes is 9%. This probability increases to 20% for highest tercile SSN. Conversely, the probability for exactly four US hurricanes is 13 and 6% for lowest and highest tercile SSN, respectively. Likewise, the probability of three or more US hurricanes occurring dwindles from 40 to 22%.

Figure 1.

Probability distribution of US hurricanes conditional on terciled August to October SSN values for years of (a) upper- and (b) lower-tercile August to October Atlantic SST anomalies. Sunspots are displayed as the averages of their respective terciles. Data span the period 1856 to 2008

In years of lowest tercile SST values among the lowest tercile August to October SSN, the probability of no US hurricanes is 17%. This probability increases to 39% for the highest tercile SSN. Note that these probabilities are greater than those for warm years, but also demonstrate the same inverse relationship for increasing SSN. Conversely, the probability for exactly four US hurricanes is 7 and 1% for the lowest and highest tercile SSN, respectively. Likewise, the probability of three or more US hurricanes dwindles from 26 to 6%. Note that the higher storm count probabilities are less than for those of the warm years. With maximum hurricane intensity dependent on SST (e.g. Emanuel, 1991; Holland, 1997; Henderson-Sellers et al., 1998), the decrease in overall US hurricane probabilities from warm SST to cold SST is expected.

Another way to consider the relationship between SSN and US hurricanes as modulated by SST is through the conditional correlation. Figure 2 shows the correlation between SSN and US hurricane counts at different percentiles of SST. The correlation (Pearson product moment) over all years is − 0.13, as indicated by the left-most point. The correlation is based on a sample size of N = 153 years. Each storm season is considered independent, and the standard error on the overall correlation estimate is 0.139, providing a 90% confidence interval on the correlation value of (−0.264, + 0.01). The confidence interval is shown by the width of the gray band.

Figure 2.

Correlation between August and October SSN and US hurricane counts. The correlations are computed at increasing percentiles of August through October averaged SST for the period 1856 to 2008. The shaded region represents the 90% confidence band at each computed value. The number of years over which the correlation is based is shown above the abscissa

The next point on the graph to the immediate right is the correlation between hurricane counts and SSN after removing the coldest 20% (20th percentile) seasons. The 20th percentile SST value constitutes an anomaly of − 0.17 °C. The correlation changes negligibly, but with reduced sample size the confidence band widens. The plot shows that the correlation does not change much for the coldest 50% of seasons. However, when 60% of the coldest seasons are removed, the correlation decreases to − 0.24 with a 90% confidence interval of (−0.429, − 0.027), based on a sample size of 61 years. The correlation continues to decrease as only the warmest years remain. With sample size decreasing accordingly, the confidence bands expand, but the relationship between the US hurricane counts and August to October SSN remains statistically significant. The strengthening negative correlation indicates that the relationship between hurricane activity and SSN is tightest when ocean temperatures are warmest. Warmer core-season SSTs provide a fertile developing ground for tropical development. As such, suppressed tropical development from high August to October SSN is most evident at this time.

4.2. Intraseasonal variability

As mentioned, the proposed sun–hurricane relationship is a result from changes in upper-level temperatures due to changes in UV radiation, thermodynamically decreasing the amount of energy for wind generation. Hood (2003) shows that the tropopause temperature response to changes in UV radiation is maximum at the 100 hPa layer around 16 km, which is commensurate with the tropical thermal tropopause and hurricane cloud height extent. Hurricane cloud heights can extend upward to the tropical tropopause layer. A warming of 0.5 K at the 100 hPa level (Hood, 2003) results in a relative temperature difference (between inflow and outflow) of only about 1%, but a relative decrease in CAPE of about 10%, according to the theory of Emanuel (1994). These thermodynamic changes decrease the relative wind speed by about 5%, which translates to a 3 m s−1 weakening of a 50 m s−1hurricane (Elsner et al., 2010). In short, increased solar activity—associated with sunspots—means more UV radiation reaching the Earth's upper atmosphere. This is corroborated by UV/upper-level temperature studies (Hood, 2003) and solar activity/ozone production increases (Angell, 1989; Calisesi and Matthes, 2007). The extra radiation contributes to warming the air aloft (via the exothermic response from ozone photodissociation), thus decreasing the temperature differential between high and low elevations that would otherwise foster vertical motion and cloud height growth. This thermodynamic explanation is in line theoretically with the observed changes in hurricane intensity concurrent with the diurnal cycle and net radiation changes (Hobgood, 1986; Gray, 1998). As mentioned, however, increased solar activity also contributes to the warming of the oceans (White et al., 1997, 1998), thus complicating the role solar variability plays in modulating hurricane activity.

Solar UV radiation at wavelengths near 200 nm is important for ozone production in the Earth's upper stratosphere. Variation in the amount of this radiation is best measured by the Mg II core-to-wing ratio (Mg II index) (Heath and Schlesinger, 1986; Lean et al., 1997; Viereck and Puga, 1999; Viereck et al., 2001) and is shown in Figure 3(a). The time series reveals the well-known 11-year Schwabe cycle and short-term fluctuations (near 27 days) caused by the asymmetric distribution of active rotating solar regions (plages and sunspots). The short-term fluctuations are especially pronounced near solar maxima (Mg II index values exceeding 0.27). Although accurate estimates of UV radiation are available only since the advent of satellite measurements, the record of sunspots dates back to the 18th century. The relationship between daily sunspot number and Mg II index is quite strong (the correlation between the two variables is 0.9), allowing us to use sunspot numbers as a reasonable proxy for UV radiation (see Figure 3(b)).

Figure 3.

Solar Mg II index values. (a) Time series of the daily solar Mg II index. The series begins on 7 November 1997 and ends on 14 October 2007 (10 578 days). Values are plotted as points. Values below the 11-year peaks indicate the solar cycle number. Horizontal axis is labelled on 1 July of the year. The inset graph shows the Mg II index values as a time series over the hurricane season of 1989. Values are connected as a curve. Breaks in the curve indicate missing values. There are 1331 missing values (12.6% of the days). (b) Scatter plot of daily sunspot number and MgII core-to-wing ratio. The correlation between the two variables is 0.9

The impact on UTT from changes in UV radiation is a fast process. Hood (2003) notes that tropical tropopause (∼15 km) temperatures vary in phase with incoming UV radiation at a zero-day phase lag, or immediately. Given such, UV radiation/UTT fluctuations are better represented on a shorter time scale than the solar influence on warming the oceans, which occurs at a monthly to multi-annual timescale. Thus, if EJ08 are correct, we should be able to detect an immediate solar influence on hurricanes by using changes in UV radiation caused by the 27-day solar rotational period.

We do this by defining solar activity during the hurricane season as anomalous if the August to October SSN is substantially different from the SSN during the months of May to July and November. Formally, let SSNanom = SSNMJJN − SSNASO, where SSNMJJN is the sunspot number averaged over May, June, July, and November, and SSNASO is the sunspot number averaged over August through October. Negative (positive) anomalies indicate that solar activity during the hurricane season is greater (less) than the solar activity during the months prior to and after the season. In general, positive anomalies arise during hurricane seasons when the Schwabe cycle is near a peak but sunspot numbers are relatively low during August, September, or October in response to the phase and intensity of the 27-day solar rotation. Negative anomalies arise during hurricane seasons when the Schwabe cycle is near a trough but sunspot numbers are relatively high during August, September, or October (Figure 4b). Thus there is a inverse relationship between total SSN and the SSN anomaly as defined here. The SSN anomaly index has a mean of − 0.80 and standard deviation of 11.37. A time series of the SSN anomaly since 1851 has been provided (Figure 4a).

Figure 4.

Seasonal sunspot number (SSN) anomalies as defined by the difference in sunspot numbers during May, June, July, November and sunspot numbers during August, September, October. (a) Time series of the anomaly and (b) scatterplot of August to October SSN versus SSN anomalies. Negative anomalies are largest for larger values of SSN. The trend has a value of − 0.08 ± 0.02 (s.e.) with dimensionless units

While the hurricane's violent vertical transport of moisture and energy overrides the ambient atmospheric temperatures, changes in temperatures at outflow levels (upper troposphere) alter the available potential energy for given vertical motion. EJ08 found a significant relationship between 150 hPa temperatures and SSN. As SSN increases, so does 150 hPa temperature. The SSN anomaly as described here offers a potential extra-terrestrial proxy for inflow and outflow temperature modification: thus, a seasonal thermodynamic profile. This is because the SSN averaged over the peripheral months of the hurricane season (May, June, July, November) captures the position of the 11-year Schwabe cycle, which is related to SST variability (Dima et al., 2005), and the SSN averaged over the core hurricane season months (August, September, October) captures the position of the 27-day cycle, which is related to upper-air variability from changes in UV radiation. By taking the difference of these two averages, we combine the effects into a single covariate, which we hypothesize is related to the probability of US hurricanes. Hurricane seasons characterized by a positive SSN anomaly (warmer inflow temperature and colder outflow temperatures) should correspond to greater hurricane activity.

Table I lists the top and bottom 10 hurricane seasons according to the value of the SSN anomaly using the years 1851–2008. The list includes the observed number of US and major US hurricanes by year along with anomaly values of other covariates which are known to be related to hurricane activity. These covariates include the NAO averaged over May and June prior to the hurricane season, the SOI averaged over the hurricane season months of June through November, and the SST averaged over June through November. The NAO is a precursor signal for hurricane steering (Elsner and Jagger, 2006) and the SOI is an indicator of the El Niño-Southern Oscillation and, therefore, wind shear and subsidence over the tropical Atlantic. Data for the SOI is available only back to 1866.

Table I. | SSN anomalies and US hurricanes: 1851–2008
+28.7 3 1 1999 +1.21 +0.43 +0.19
+27.9 1 0 1984 − 1.01 − 0.22 − 0.26
+26.6 2 1 1938 +0.31 +1.19 +0.29
+25.3 4 1 1906 − 1.72 +0.99 − 0.05
+21.7 4 2 1916 − 1.88 +1.06 − 0.06
+20.8 1 1 1885 +0.96 − 1.23 +0.01
+20.1 2 1 1894 +0.7 − 0.29 − 0.27
+19.0 2 1 1929 +0.31 +0.3 − 0.12
+18.3 3 1 1871 − 1.94 +0.23 +0.01
+17.9 3 2 1950 − 0.24 +1.59 +0.01
+22.6 2.5 1.1 1927 −0.33 +0.4 −0.03
  1. Years are listed according to the value of the ten most negative SSN anomalies (top) and ten most positive (bottom). Corresponding columns include the number of US hurricanes (US), major US hurricanes (MUS), the NAO as a May to June average anomaly, the SOI as a June to November average anomaly, and SST as a June to November average anomaly. Averages over the 10 years are given in bold.

−37.9 0 0 1981 −0.76 0.41 −0.07
−27.4 0 0 1990 −0.39 −0.31 +0.07
−26.9 6 0 1858 +0.15 −0.17
−26.4 1 0 1909 +1.38 +0.58 −0.06
−25.0 1 1 1957 −0.82 −0.6 +0.14
−24.8 4 1 1880 −0.67 +0.75 +0.13
−22.8 0 0 2001 −0.71 −0.18 +0.25
−21.7 1 1 1958 −0.17 −0.14 +0.21
−20.6 0 0 1978 +0.26 +0.01 −0.19
−25.4 1.4 0.3 1947 +0.12 +0.16 +0.02

Of the top ten positive anomaly years, 2 years featured four US hurricanes and 3 years featured three. There was at least one US hurricane in each of the top ten anomalous years. In contrast, four of the bottom 10 years had no US hurricanes and 3 years had only one. The mean US hurricane rate for positive SSN anomaly years is 37% higher than the overall mean US rate (1.82 hurricanes per year). For negative differences, mean US counts were 56% lower. Mean major US (MUS) hurricanes demonstrated a 72% increase (positive) and a 53% decrease (negative), respectively, in departure from mean seasonal MUS hurricanes (0.64 hurricanes per year). Mean values of the SST and SOI anomalies are clearly not able to explain the difference in hurricane rates; however, the NAO might be a confounding variable. This is examined next in the context of a multivariate regression model.

4.3. A seasonal model incorporating the SSN anomaly

To examine the intra-seasonal SSN anomaly as a predictor for US hurricanes, we need a multivariate setting. This is achieved through the development of a regression model. Separate models are developed for hurricanes and major hurricanes that make landfall in the United States. The Poisson regression is a form of statistical regression used to model count data, such as the annual number of US hurricanes. The Poisson regression assumes that the response variable H (annual number of US hurricanes) has a Poisson distribution with a single parameter λ, which is the annual rate of US hurricanes. The logarithm of the annual rate is modelled using a linear combination of covariates. The model coefficients are determined using the method of maximum likelihood. The model covariates include May to June NAO index as an indicator of hurricane steering, the June to November SOI as an indicator of Atlantic basin wind shear and subsidence, June to November SST as an indicator of ocean heat content, and the SSN anomaly as defined above. With the exception of the SSN anomaly, the covariates are identical to those used in Elsner and Jagger (2006) for predicting US hurricane activity.

The NAO index is the sea-level pressure difference between the permanent Icelandic Low and semi-permanent Azores High. An inverse relationship between the NAO index and US hurricanes was found in Elsner and Kocher (2000). When the Azores High is located farther to the north and east and is stronger (i.e. high NAO index values), hurricanes that form over the central North Atlantic tend to get steered away from the United States. The SOI is another sea-level pressure difference (Tahiti minus Darwin, Australia) that has been shown to vary in phase with North Atlantic hurricane activity (Gray, 1984).

Low values of the SOI correspond with warm eastern equatorial Pacific SST (i.e. El Niño conditions). This warm water tends to shift the focus of Pacific deep tropical convection eastward, leading to sheering winds and subsidence over the North Atlantic basin. North Atlantic hurricane activity also varies in phase with SST. Warm waters provide a breeding ground for genesis and intensification of tropical cyclones, with temperatures below 26 °C generally too cool for supporting organized tropical deep convection (Gray, 1968).

Table II lists model parameters and analysis of deviance for a model with the number of US hurricanes as the response variable from 1866 to 2008. The coefficient on the SSN anomaly term indicates a positive relationship with hurricane frequency, consistent with the relationship proposed from the analysis of Table I. Model covariates are added sequentially to the model. Model parameters and analysis of deviance for the first three covariates were originally shown to be statistically significant in EJ08. The SSN anomaly is determined to be the second most important variable (based on the magnitude of deviance) toward explaining the US hurricanes. The model for major US hurricanes is similar, but here the SSN anomaly term is the most important variable.

Table II. | Poisson regression of US hurricanes
Term Estimate Dev. Res. df Res. Dev. p-Value
US hurricanes
  1. Regression parameters using all US hurricanes as the response (top) and all major US hurricanes as the response (bottom) for 1866–2008. The columns include the parameter estimate, the deviance (Dev.), the residual degrees of freedom (Res. df), the residual deviance (Res. Dev.) and the p-value. The p-values are based on F-tests for differences in group means. All p-values are less than 0.1, indicating the significance of the term to the model after accounting for the terms already in the model.

NULL 0.498 142 180.756  
NAO −0.176 12.166 141 168.589 0.005
SOI +0.157 6.254 140 162.336 0.022
SST +0.665 4.186 139 158.150 0.034
SSNanom +0.028 6.447 138 151.703 0.011
    Major US hurricanes      
NULL -0.604   142 152.929  
NAO −0.190 5.654 141 147.275 0.073
SOI +0.256 5.936 140 141.338 0.022
SST +1.144 4.424 139 136.914 0.030
SSNanom +0.047 6.624 138 130.290 0.010

Note that the SSN anomaly covariate is significant after accounting for the NAO. In a general circulation model, Shindell et al. (2001) find a shift to low index values of the NAO as a response to reduced solar irradiance. This would contribute to an increase in US hurricane probability, as the paths they take are directed through the Caribbean Sea. In fact, the correlation between June values of SSN and the NAO (1866–2008) is + 0.14 (p-value = 0.1), suggesting that some of the solar influence on hurricanes might be through solar-forced changes to the NAO. More work on this area is needed.

Model adequacy is checked by examining the residual deviance after all terms are included. For the hurricane model, the residual deviance is 151.7 with 138 degrees of freedom (df). The p-value as evidence in favour of model adequacy from a χ2 distribution with this quantile value and this df is 0.2, indicating no strong evidence against model adequacy. The p-value on the residual deviance of the major hurricane model is 0.7, again providing no evidence against model adequacy.

Figure 5 illustrates the model-predicted changes in US hurricane frequency with respect to the extremes of SSN anomalies while assuming neutral conditions for the NAO, SST, and SOI. A zero-count US hurricane season is more than 3 times as likely to occur in a low SSN anomaly season (32%) than high (9%). The chances of at least one major hurricane affecting the United States in the lowest and highest SSN anomaly season are 10 and 39%, respectively.

Figure 5.

Predicted probability distribution of US hurricanes. Predictions are made using a Poisson regression for values of SSN anomalies corresponding to the 1st and 99th percentiles while keeping the other three covariates at neutral values. The model is based on data from the period 1866–2008

5. The sun–hurricane relationship prior to 1851

Above, we showed that an index that quantifies a portion of the intra-hurricane season sunspot variability is quantitatively related to US hurricane and major hurricane frequency controlling for the other climate variables known to affect hurricanes. The SSN anomaly index captures, to some extent, the combination of low frequency variation associated with the 11-year Schwabe cycle and the 27-day solar rotation. This is important as it allows us to examine the relationship farther back in time with the availability of long records of sunspot numbers and old records of hurricanes over the western part of the North Atlantic basin. In this section, we consider the sun–hurricane relationship back to 1749, which is the start of accurate monthly sunspot counts with the International Sunspot Number, and make use of a portion of the Chenoweth (2006) Atlantic hurricane archive.

Figure 6 shows the time series of US counts since 1749 where the counts from the Chenoweth (2006) archive are appended to the HURDAT counts used in the previous section. Hurricane intensity entries in the archive listing a US mainland state or region are qualified as ‘USA hurricane’.

Figure 6.

Annual US hurricane counts for the periods (a) 1749–1850 and (b) 1851–2008. Data from 1851 through 2008 are from HURDAT and the data prior to 1851 are from the Chenoweth (2006) historical hurricane archive

Historical archives are less comprehensive than modern records. So differences in hurricane rates between the modern (HURDAT) and historical (Chenoweth archive) storm records can largely be explained by differences in completeness of the records. Given the nature of the collection of pre-1851 observations, knowledge of cyclone location and occurrence was a function of shipping traffic path frequency and meteorologically attuned populations. As such, for each year there is probability of at least one hurricane making landfall which is not in the record and this probability is almost certainly larger at the beginning of the record than towards the end accounting for, at least partially, the upward trend over this period. Thus we assume that the collection of hurricane counts prior to 1851 is an under-representation of the actual counts.

This under-count notwithstanding, here we are interested in the variability in available counts from one year to the next. We can assume that this variability is largely unaffected by trends caused by missing cyclones. In fact, the autocorrelation function computed using the counts from 1749 to 1850 is similar to that computed using the counts from 1851 to 2008. Moreover, it is certainly not likely that observational limitations correlate with the solar cycle.

Overlapping portions of the Chenoweth archive and SSN data provide a 102-year segment from which we can further examine the sun–hurricane relationship. The two busiest years in the series are the 1835 and 1837 seasons, with each reporting four hurricanes impacting the United States. There are 73 seasons that record at least one US hurricane, and the mean number of recorded hurricanes is 1.04 per year, giving an annual probability of 28% of at least one hurricane occurrence. This compares with the modern record mean (1.79 hurricanes per year) and annual probability of at least one hurricane occurrence (53%).

Table III lists the top and bottom ten hurricane seasons according to the value of the SSN anomaly using the years 1749–1850. The list includes the estimated number of US hurricanes by year. No direct observations for Atlantic SSTs, the ENSO, or the NAO are available for these early years. Of the top ten positive anomaly years, one year featured four recorded US hurricanes and five years featured two US hurricanes. With the exception of 1774, there was at least one US hurricane in each of the top ten anomalous years. In contrast, eight of the bottom ten years had fewer than two US hurricanes, although 1835 featured four.

Table III. | SSN anomalies and US hurricanes: 1749–1850
SSN anomaly USA Year
+37.6 2 1781
+32.9 1 1749
+26.6 1 1771
+24.9 2 1794
+23.4 2 1838
+22.5 2 1778
+22.1 1 1789
+19.5 2 1752
+19.3 0 1774
+16.9 4 1837
+24.5 1.7 1786
SSN Anomaly USA Year
  1. Years are listed according to the value of the ten most negative SSN anomalies (top) and ten most positive SSN anomalies (bottom). Corresponding columns include the number of US hurricanes (USA). Averages over the 10 years are given in bold.

−72.7 0 1847
−52.8 1 1839
−33.8 1 1769
−25.9 4 1835
−24.6 2 1787
−20.3 1 1759
−16.9 0 1786
−16.6 1 1850
−15.0 1 1824
−14.6 1 1846
−29.3 1.2 1814

The mean US hurricane rate is 42% higher in the ten positive SSN anomaly years compared to the ten negative anomaly years (the 1851–2008 mean difference is 79%, respectively). However, note that the mean year for the positive anomalies is 1786, 28 years earlier than the mean year for the negative anomalies (1814). Since the idea is for fewer missed hurricanes as time goes by, the actual difference in hurricane rates is probably greater than shown here and closer to what is found over the period 1851–2008. Thus, despite the inherent limitations in the older hurricane data, there is a detectable solar signal that mirrors the signal found in the more recent records, featuring a higher probability of US hurricanes when the SSN anomaly is positive and large.

6. Summary and conclusions

Here we examined the evidence for a sun–hurricane relationship for hurricanes affecting the United States, which was identified recently by EJ08. We first looked at seasonal activity and noted a consistent inverse relationship between the probability of a US hurricane and SSN. The relationship occurs for both warm and cold years as defined by Atlantic SSTs. Since it has been argued that the sun–hurricane relationship arises from changes in upper tropospheric temperature associated with variations in the UV radiation (Elsner et al., 2010), next we showed the strong correlation between SSN and a core-to-wing ratio as a satellite-derived measure of the UV flux.

Since the SST responds on a slower timescale to changes in total solar irradiance, we derived a SSN anomaly that captures a portion of the intra-hurricane season variability in solar activity by subtracting the SSN averaged over August through October from the SSN averaged over May, June, July, and November. The SSN anomaly is used to divide the hurricane record from 1851 to 2008 into top and bottom years with respect to the 20 most anomalous values. As expected, years with positive SSN anomalies featuring high peripheral month sunspot numbers but low in-season numbers have, on average, significantly more (79%) US hurricanes. This relationship was checked over all years by controlling for SST, the ENSO cycle, and the NAO in a multivariate Poisson regression model. The SSN anomaly was shown to be statistically significant in models for US hurricanes and US major hurricanes after accounting for the other climate variables.

The evidence for a sun–hurricane relationship was further bolstered by showing that a similar relationship between the SSN anomaly and US hurricanes (years of high SSN anomaly have more US hurricanes) is detectable in an archive of Atlantic hurricanes dating back to 1749. This lends additional support to the contention that solar variability influences hurricane activity and provides support for the utility of older, less reliable, hurricane records for examining hurricane–climate relationships. Consequently, we feel that this work makes an important new contribution by showing that the sun–hurricane relationship, discussed in EJ08 using mostly 20th century data, does not disappear when examined by another 100 years of data. The work will be extended by examining whether the relationship can be seen in other tropical cyclone basins.

Our thermodynamic hypothesis explaining the observed correlation between hurricanes and the solar activity is likely incomplete. As mentioned, increased solar irradiance tends to trend the NAO towards positive anomalies, which reduces the chance that a hurricane will reach the Unites States. Also, under an active sun the stratosphere warms unevenly, with the most pronounced warming occurring at lower latitudes. As a result, stratospheric winds are altered, which could end up changing the strength of tropical cyclones (van Loon et al., 2004; Meehl et al., 2009; van Loon et al., 2007; van Loon and Shea, 1999). Our hunch is that circulation changes could influence weaker tropical cyclones but the intensity of the stronger hurricanes is influenced more by the thermodynamic changes above and below the hurricane. More work in this area is needed.

At the time of writing, the 2010 hurricane season is set to occur during the onset stage of Solar Cycle 24, with a running average estimate of about 30 sunspots over the duration of the hurricane season. Using monthly sunspot predictions for from NOAA/Space Weather Prediction Center, the 2010 hurricane season could feature an SSN anomaly of approximately − 5, or around the 32nd percentile of SSN anomaly activity. Utilizing a model similar to that used in Figure 5 and assuming the other three climate values at neutral values (they are clearly not), an SSN anomaly of − 5 corresponds to a 46% chance of two or more US hurricanes, 8% lower than the climatological average (54%) for two or more US hurricanes.


We thank Thomas H. Jagger for help with some of the statistics. Thanks are extended to the SIDC for the sunspot data and to the NHC for HURDAT. All statistical modelling was done using R (R Development Core Team, 2009). The work is supported by the US National Science Foundation (ATM-0738172) and the Risk Prediction Initiative (RPI-08-02-002) of the Bermuda Institute for Ocean Sciences. Interpretations and opinions expressed here are those of the authors and do not necessarily reflect those of the funding agencies.

Link to comment

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
  • Create New...

Important Information

Our Privacy Policy can be found here: Privacy Policy. We have placed cookies on your device to help make this website better. You can adjust your cookie settings, otherwise we'll assume you're okay to continue..